File System Design Using Large Memories

Abstract:

Over the last few years tremendous strides have been made in CPU performance, but without corresponding strides in I/O performance. Consequently, operating systems must be redesigned to minimize the impact of the I/O bottleneck. Memory is becoming cheaper and may be effectively used to improve I/O performance, but more sophisticated methods than simple demand caching may exist to make caching more effective. Analysis of file use in current file systems has shown the existence of file access patterns which may be exploited to provide better I/O performance. Using experimental data, we show what implications this has for file system design. We outline the design of the iPcress File System which
uses both a large disk cache and other techniques to improve file system performance. Preliminary performance numbers for iPcress are also presented.